PASSIVE TORPEDO DETECTION/CLASSIFICATION ALGORITHM DEVELOPMENT (U)

THE DEVELOPMENT OF ALGORITHMS FOR THE TIMELY PASSIVE DETECTION AND CLASSIFICATION OF TORPEDOES USING THEIR ACOUSTIC RADIATED SIGNATURES IS A HIGH PRIORITY OF BOTH THE SURFACE SHIP AND SUBMARINE TORPEDO-DEFENSE PROGRAMS. IN THE LAST SEVERAL YEARS, 0RINC0N HAS EMPLOYED NEURAL NETWORK TECHNOLOGY IN THE DETECTION AND CLASSIFICATION OF A NUMBER OF UNDERWATER ACOUSTIC SIGNALS OF INTEREST TO THE NAVY. NEURAL NETWORKS ARE ROBUST AND DO NOT DEPEND ON ANY ASSUMPTIONS ABOUT NOISE OR SIGNAL DISTRIBUTION. AS A RESULT, THEY CAN BE EASILY MODIFIED TO ACCOMMODATE CHANGING THREAT SIGNAL TYPES WITH NO MAJOR SYSTEM CHANGES. HOWEVER, IN A COMPLEX ACOUSTIC ENVIRONMENT INVOLVING OWNSHIP, THE THREATSUBMARINE, ONE OR MORE TORPEDOES, THE FALSE ALARM AND/OR INCORRECT CLASSIFICATION RATES WITH JUST A NEURAL NETWORK MAY BE GREATER THAN DESIRED. TO FURTHER REDUCE THE FALSE ALARM RATE, INCREASE THE NUMBER OF CORRECT CLASSIFICATIONS, AND PROVIDE A HIGHER LEVEL SCENE ANALYSIS,AN EXPERT SYSTEM CAN BE UTILIZED EITHER AS A BACK-END PROCESS OR AN ADJUNCT TO THE NEURAL NETWORK. THE PRIMARY PURPOSE OF THIS PHASE I EFFORT IS TO DEMONSTRATE THE APPLICABILITY OF THIS APPROACH BY DEVELOPING A PROTOTYPE SYSTEM COMBINING AN ADVANCED ADEC-LIKE LINE TRACKING ALGORITHM FOR DETECTION WITH A GRAM ANALYSIS AND MONITORING EXPERT SYSTEM FOR CLASSIFICATION AND FALSE ALARM REDUCTION. THIS PROTOTYPE SYSTEM WOULD BE EVALUATED USING A VARIETY TEST DATA SUPPLIED BY THE NAVY.